Scribd
Upload
37 views6 pages

Project Plan For RAG Algorithm

Project CSEC aims to develop a model for retrieving project information based on natural language queries related to structural features and unique design elements. The project outlines four model versions, with the latest focusing on a hybrid approach that combines vector embeddings and GPT reasoning for improved accuracy. Next steps include fine-tuning the model, expanding the dataset, and making the model accessible through a user-friendly interface.

Uploaded by

Mohammed Salman
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
37 views6 pages

Project Plan For RAG Algorithm

Project CSEC aims to develop a model for retrieving project information based on natural language queries related to structural features and unique design elements. The project outlines four model versions, with the latest focusing on a hybrid approach that combines vector embeddings and GPT reasoning for improved accuracy. Next steps include fine-tuning the model, expanding the dataset, and making the model accessible through a user-friendly interface.

Uploaded by

Mohammed Salman
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Project CSEC

A Model to retrieve projects information’s (technical drawings/relevant documents) based on


natural language input — such as structural features or unique design elements

For example:

1. Structural Criteria: “List buildings with steel roofs and more than 2 stories.”

2. Unique Features: “Which projects have circular staircases?”

Extracted Project Dataset:

Model’s Data Structure:


Prototype Model’s

1. Model V1 – Text Similarity (Keyword Match)


- simple string-matching using Python & CSV

- Downsides: It only finds results if the query matches the


text exactly (no meaning or relationships understanding)
2. Model V2 – Vector Embedding Search (Semantic)
- use Sentence Transformers + FAISS (facebook ai similarity
check)
- Embeddings capture meaning but ignore structured
relationships
- Example: It might find “pile foundation” similar to “raft
foundation” just because they occur in similar contexts not
because they’re technically related.
- Can't enforce filters like: "only projects with 2+ stories
AND circular staircase AND pile foundation".
3. Model V3 – Hybrid (Embedding + GPT Reasoning)
- use vector embedding for narrowing candidates, then GPT-
4 for filtering & reasoning
- Operational Cost
- Need improvement for higher accuracy.
4. Model V4 – Hybrid model with domain specific fine
tunings
- Highly Accurate + Domain-Aware Responses

Next Steps

1. Fine-Tune the Model (for Higher Accuracy)


- Implementing model V4
Goal: Teach the model domain-specific knowledge using
examples (like project queries & expected outputs and their
relationships)

2. Improve with New Data + Set Up Relational DB


Goal: Expand and organize your data
(drawings/projects/features) to support scalable search &
retrieval.

3. Make the Model Accessible to Others (Web App or)


- Design UI
- Connect UI to backend
- Create a cloud/server DB and deploy the system

You might also like